Conda environment and Docker Image

[cauachagas]

• Dependencies from Anaconda.org
• Docker image (miniconda3)

I tested the Docker image with mpich only, but I think it works with openmpi too.

[scivision]

This has the issue of when using a system where MKL libraries are installed from Intel directly, CMake won't find them.
That is, if a system has MKL libraries installed directly, and has the conda environment activated but without Conda MKL, CMake doesn't find the libraries at all.
I think it needs to check if MKLROOT is defined already, and if so, assume that's what the user wants.

[scivision]

This MKLROOT <= conda_prefix assumption actually breaks any system that has conda activated but does not have MKL installed in conda, regardless of whether Intel MKL libraries are installed directly or not.

The Find*.cmake modules here assume that if MKLROOT is defined, the user must be intending for MKL to be used, and will fail if MKL is not found.

The problem might be approached by instead altering the Find*.cmake logic to look inside conda_prefix to see if MKL is in there, and fall back to generic libraries if not.

For now the simplest approach / workaround from the package's perspective is to omit this change and have the user define environment MKLROOT themselves, or with cmake -DMKLROOT=$CONDA_PREFIX.

[cauachagas]

-DMKLROOT=$CONDA_PREFIX ... So simple ...

I have no experience with CMake, but a lot of projects currently use CMake, so I decided to learn and found the repository when searching for Pardiso + Cmake on Google.

Some tips posted at https://www.scivision.dev are being useful for my research. Thank you for sharing.